[Bug c/37103] New: possible integer codegen bug

[Bug c/37103] New: possible integer codegen bug

[regehr at cs dot utah dot edu]

This is an interesting one...

Compiling the code below at all common optimization levels, gcc r139046
generates code that prints hello, as does 4.3.1.  Pre-4.3 versions of gcc emit
code that prints nothing.  I'm pretty sure that the older versions are correct.

#include <stdio.h>

int func_72 (void);
int func_72 (void)
{
  printf ("hello\n");
  return 0;
}

void func_58 (unsigned short p_65);
void func_58 (unsigned short p_65)
{
  char g_99 = -1;
  (p_65 != g_99) || (func_72 ());
}

int main (void)
{
  func_58 (-1);
  return 0;
}


-- 
           Summary: possible integer codegen bug
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: regehr at cs dot utah dot edu
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

Re: [Bug c/37103] New: possible integer codegen bug

[Andrew Thomas Pinski]

Note for most targets not printing is correct as char is signed by  
default but for most powerpc targets the opposite is true. You should  
have explicted included signed for g_99.

Sent from my iPhone

On Aug 12, 2008, at 22:10, "regehr at cs dot utah dot edu" <gcc-bugzilla@gcc.gnu.org 
 > wrote:

> This is an interesting one...
>
> Compiling the code below at all common optimization levels, gcc  
> r139046
> generates code that prints hello, as does 4.3.1.  Pre-4.3 versions  
> of gcc emit
> code that prints nothing.  I'm pretty sure that the older versions  
> are correct.
>
> #include <stdio.h>
>
> int func_72 (void);
> int func_72 (void)
> {
>  printf ("hello\n");
>  return 0;
> }
>
> void func_58 (unsigned short p_65);
> void func_58 (unsigned short p_65)
> {
>  char g_99 = -1;
>  (p_65 != g_99) || (func_72 ());
> }
>
> int main (void)
> {
>  func_58 (-1);
>  return 0;
> }
>
>
> -- 
>           Summary: possible integer codegen bug
>           Product: gcc
>           Version: 4.4.0
>            Status: UNCONFIRMED
>          Severity: normal
>          Priority: P3
>         Component: c
>        AssignedTo: unassigned at gcc dot gnu dot org
>        ReportedBy: regehr at cs dot utah dot edu
> GCC build triplet: i686-pc-linux-gnu
>  GCC host triplet: i686-pc-linux-gnu
> GCC target triplet: i686-pc-linux-gnu
>
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103
>

[Bug c/37103] possible integer codegen bug

[pinskia at gmail dot com]

------- Comment #1 from pinskia at gmail dot com  2008-08-13 05:44 -------
Subject: Re:   New: possible integer codegen bug

Note for most targets not printing is correct as char is signed by  
default but for most powerpc targets the opposite is true. You should  
have explicted included signed for g_99.

Sent from my iPhone

On Aug 12, 2008, at 22:10, "regehr at cs dot utah dot edu"
<gcc-bugzilla@gcc.gnu.org 
 > wrote:

> This is an interesting one...
>
> Compiling the code below at all common optimization levels, gcc  
> r139046
> generates code that prints hello, as does 4.3.1.  Pre-4.3 versions  
> of gcc emit
> code that prints nothing.  I'm pretty sure that the older versions  
> are correct.
>
> #include <stdio.h>
>
> int func_72 (void);
> int func_72 (void)
> {
>  printf ("hello\n");
>  return 0;
> }
>
> void func_58 (unsigned short p_65);
> void func_58 (unsigned short p_65)
> {
>  char g_99 = -1;
>  (p_65 != g_99) || (func_72 ());
> }
>
> int main (void)
> {
>  func_58 (-1);
>  return 0;
> }
>
>
> -- 
>           Summary: possible integer codegen bug
>           Product: gcc
>           Version: 4.4.0
>            Status: UNCONFIRMED
>          Severity: normal
>          Priority: P3
>         Component: c
>        AssignedTo: unassigned at gcc dot gnu dot org
>        ReportedBy: regehr at cs dot utah dot edu
> GCC build triplet: i686-pc-linux-gnu
>  GCC host triplet: i686-pc-linux-gnu
> GCC target triplet: i686-pc-linux-gnu
>
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103
>


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2008-08-13 09:16 -------
The C standard specifies that the comparison is done in type int which means
that for both signed and unsigned char

  (char)-1 != (unsigned short)-1

should evaluate to false.  Testcase:

extern void abort (void);

void func_1 (unsigned short p_65)
{
  signed char g_99 = -1;
  if (p_65 == g_99)
    abort ();
}

void func_2 (unsigned short p_65)
{
  unsigned char g_99 = -1;
  if (p_65 == g_99)
    abort ();
}

int main (void)
{
  func_1 (-1);
  func_2 (-1);
  return 0;
}

This looks like yet another folding bug as we end up with the comparison
narrowed.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
          Component|c                           |middle-end
     Ever Confirmed|0                           |1
           Keywords|                            |wrong-code
      Known to fail|                            |4.3.1 4.4.0
      Known to work|                            |4.2.4
   Last reconfirmed|0000-00-00 00:00:00         |2008-08-13 09:16:17
               date|                            |
            Summary|possible integer codegen bug|[4.3/4.4 Regression]
                   |                            |possible integer codegen bug
   Target Milestone|---                         |4.3.2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[jakub at gcc dot gnu dot org]

------- Comment #3 from jakub at gcc dot gnu dot org  2008-08-13 13:37 -------
I bet 132269 or 134108 are the cause, will test later.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[jakub at gcc dot gnu dot org]

------- Comment #4 from jakub at gcc dot gnu dot org  2008-08-13 18:30 -------
Verified the regression was introduced by PR35163 fix.
http://gcc.gnu.org/ml/gcc-patches/2008-04/msg00720.html
should have been IMNSHO applied for shorter_type bigger or equal than
arg1_unw's type, not just when they are equal.  If shorter_type is bigger and
the signs are different, then the bits above shorter_type bits might be all 1's
in one case and all 0's in the second case.

Patch I'll be regtesting:

2008-08-13  Jakub Jelinek  <jakub@redhat.com>

PR middle-end/37103
* fold-const.c (fold_widened_comparison): Do not allow
sign changes that change the result even if shorter type
is wider than arg1_unw's type.

* gcc.c-torture/execute/20080813-1.c: New test.

--- gcc/fold-const.c.jj2008-08-13 19:46:11.000000000 +0200
+++ gcc/fold-const.c2008-08-13 20:18:21.000000000 +0200
@@ -6733,10 +6733,8 @@ fold_widened_comparison (enum tree_code 
   if ((code == EQ_EXPR || code == NE_EXPR
        || TYPE_UNSIGNED (TREE_TYPE (arg0)) == TYPE_UNSIGNED (shorter_type))
       && (TREE_TYPE (arg1_unw) == shorter_type
-  || (TYPE_PRECISION (shorter_type)
-      > TYPE_PRECISION (TREE_TYPE (arg1_unw)))
   || ((TYPE_PRECISION (shorter_type)
-       == TYPE_PRECISION (TREE_TYPE (arg1_unw)))
+       >= TYPE_PRECISION (TREE_TYPE (arg1_unw)))
       && (TYPE_UNSIGNED (shorter_type)
   == TYPE_UNSIGNED (TREE_TYPE (arg1_unw))))
   || (TREE_CODE (arg1_unw) == INTEGER_CST
--- gcc/testsuite/gcc.c-torture/execute/20080813-1.c.jj2008-08-13
20:22:56.000000000 +0200
+++ gcc/testsuite/gcc.c-torture/execute/20080813-1.c2008-08-13
20:22:10.000000000 +0200
@@ -0,0 +1,30 @@
+/* PR middle-end/37103 */
+
+extern void abort (void);
+
+void
+foo (unsigned short x)
+{
+  signed char y = -1;
+  if (x == y)
+    abort ();
+}
+
+void
+bar (unsigned short x)
+{
+  unsigned char y = -1;
+  if (x == y)
+    abort ();
+}
+
+int
+main (void)
+{
+  if (sizeof (int) == sizeof (short))
+    return 0;
+  foo (-1);
+  if (sizeof (short) > 1)
+    bar (-1);
+  return 0;
+}


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |jakub at gcc dot gnu dot org
                   |dot org                     |
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2008-08-13 09:16:17         |2008-08-13 18:30:25
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[rguenther at suse dot de]

------- Comment #5 from rguenther at suse dot de  2008-08-13 21:05 -------
Subject: Re:  [4.3/4.4 Regression] possible integer
 codegen bug

On Wed, 13 Aug 2008, jakub at gcc dot gnu dot org wrote:

> ------- Comment #4 from jakub at gcc dot gnu dot org  2008-08-13 18:30 -------
> Verified the regression was introduced by PR35163 fix.
> http://gcc.gnu.org/ml/gcc-patches/2008-04/msg00720.html
> should have been IMNSHO applied for shorter_type bigger or equal than
> arg1_unw's type, not just when they are equal.  If shorter_type is bigger and
> the signs are different, then the bits above shorter_type bits might be all 1's
> in one case and all 0's in the second case.

Yeah, you are right.  Patch is pre-approved.

Richard.

> Patch I'll be regtesting:
> 
> 2008-08-13  Jakub Jelinek  <jakub@redhat.com>
> 
> PR middle-end/37103
> * fold-const.c (fold_widened_comparison): Do not allow
> sign changes that change the result even if shorter type
> is wider than arg1_unw's type.
> 
> * gcc.c-torture/execute/20080813-1.c: New test.
> 
> --- gcc/fold-const.c.jj2008-08-13 19:46:11.000000000 +0200
> +++ gcc/fold-const.c2008-08-13 20:18:21.000000000 +0200
> @@ -6733,10 +6733,8 @@ fold_widened_comparison (enum tree_code 
>    if ((code == EQ_EXPR || code == NE_EXPR
>         || TYPE_UNSIGNED (TREE_TYPE (arg0)) == TYPE_UNSIGNED (shorter_type))
>        && (TREE_TYPE (arg1_unw) == shorter_type
> -  || (TYPE_PRECISION (shorter_type)
> -      > TYPE_PRECISION (TREE_TYPE (arg1_unw)))
>    || ((TYPE_PRECISION (shorter_type)
> -       == TYPE_PRECISION (TREE_TYPE (arg1_unw)))
> +       >= TYPE_PRECISION (TREE_TYPE (arg1_unw)))
>        && (TYPE_UNSIGNED (shorter_type)
>    == TYPE_UNSIGNED (TREE_TYPE (arg1_unw))))
>    || (TREE_CODE (arg1_unw) == INTEGER_CST
> --- gcc/testsuite/gcc.c-torture/execute/20080813-1.c.jj2008-08-13
> 20:22:56.000000000 +0200
> +++ gcc/testsuite/gcc.c-torture/execute/20080813-1.c2008-08-13
> 20:22:10.000000000 +0200
> @@ -0,0 +1,30 @@
> +/* PR middle-end/37103 */
> +
> +extern void abort (void);
> +
> +void
> +foo (unsigned short x)
> +{
> +  signed char y = -1;
> +  if (x == y)
> +    abort ();
> +}
> +
> +void
> +bar (unsigned short x)
> +{
> +  unsigned char y = -1;
> +  if (x == y)
> +    abort ();
> +}
> +
> +int
> +main (void)
> +{
> +  if (sizeof (int) == sizeof (short))
> +    return 0;
> +  foo (-1);
> +  if (sizeof (short) > 1)
> +    bar (-1);
> +  return 0;
> +}
> 
> 
> 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[regehr at cs dot utah dot edu]

------- Comment #6 from regehr at cs dot utah dot edu  2008-08-13 22:52 -------
Created an attachment (id=16067)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16067&action=view)
potential test program for gcc testsuite

I wrote a little script to make a C program that exhausts the possible integer
comparisons for -1, 0, 1.  Interested in adding this to the test suite?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[regehr at cs dot utah dot edu]

------- Comment #7 from regehr at cs dot utah dot edu  2008-08-13 22:57 -------
(In reply to comment #6)
> Created an attachment (id=16067)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16067&action=view) [edit]
> potential test program for gcc testsuite
> 
> I wrote a little script to make a C program that exhausts the possible integer
> comparisons for -1, 0, 1.  Interested in adding this to the test suite?

The answers are only valid for ia32 of course... if a platform-independent
version of this would be interesting, let me know.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[jakub at gcc dot gnu dot org]

------- Comment #8 from jakub at gcc dot gnu dot org  2008-08-14 09:04 -------
Subject: Bug 37103

Author: jakub
Date: Thu Aug 14 09:02:46 2008
New Revision: 139093

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=139093
Log:
        PR middle-end/37103
        * fold-const.c (fold_widened_comparison): Do not allow
        sign changes that change the result even if shorter type
        is wider than arg1_unw's type.

        * gcc.c-torture/execute/20080813-1.c: New test.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/20080813-1.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/fold-const.c
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[jakub at gcc dot gnu dot org]

------- Comment #9 from jakub at gcc dot gnu dot org  2008-08-14 09:12 -------
Subject: Bug 37103

Author: jakub
Date: Thu Aug 14 09:11:03 2008
New Revision: 139094

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=139094
Log:
        PR middle-end/37103
        * fold-const.c (fold_widened_comparison): Do not allow
        sign changes that change the result even if shorter type
        is wider than arg1_unw's type.

        * gcc.c-torture/execute/20080813-1.c: New test.

Added:
    branches/gcc-4_3-branch/gcc/testsuite/gcc.c-torture/execute/20080813-1.c
Modified:
    branches/gcc-4_3-branch/gcc/ChangeLog
    branches/gcc-4_3-branch/gcc/fold-const.c
    branches/gcc-4_3-branch/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103

[Bug middle-end/37103] [4.3/4.4 Regression] possible integer codegen bug

[jakub at gcc dot gnu dot org]

------- Comment #10 from jakub at gcc dot gnu dot org  2008-08-14 09:20 -------
Fixed in CVS.
Regarding your testcase, I think it is too big, but its size could be very well
decreased just by using preprocessor extensively.  That said, I'm not sure if
some embedded targets won't be upset about 5401 functions.


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37103